Crash analysis for urban road networks: Semi-parametric spatial negative binomial models with monotonic constraints

Area of Science:

• Transportation Engineering
• Spatial Statistics
• Econometrics

Background:

• Spatial heterogeneity is crucial in crash frequency modeling.
• Existing models have limitations in exposure variable flexibility and computational efficiency.
• Addressing these limitations is key to advancing road safety analysis.

Purpose of the Study:

• To develop a semi-parametric spatial count model with flexible exposure specifications and efficient Bayesian estimation.
• To overcome limitations of current spatial methodologies in crash modeling.
• To provide a more accurate framework for understanding and mitigating road crash risks.

Main Methods:

• Developed a semi-parametric spatial count model incorporating flexible exposure specifications and spatial dependence.
• Employed data-augmented Gibbs sampling for computationally efficient Bayesian estimation using conjugate forms.
• Validated the model through simulations and empirical evaluations on urban road networks in Houston and Dallas.

Main Results:

• The proposed model demonstrated significant performance advantages over conventional approaches.
• Analysis revealed complex, segment-varying relationships between annual average daily traffic (AADT) and crash frequency, with city-specific patterns.
• Roadway design elements (lane number, width, median design) and spatial heterogeneity were identified as significant crash risk factors.

Conclusions:

• The semi-parametric spatial count model offers a more flexible and computationally efficient approach to road safety analysis.
• Understanding spatial heterogeneity and complex exposure-risk relationships is vital for effective traffic safety interventions.
• The framework enables transportation agencies to prioritize safety improvements by precisely quantifying crash risk factors.